An LLM judge cannot be a build gate, and it is not about the cost

Almost every RAG evaluation metric on offer needs a language model to produce it. Faithfulness,...